| Resumo: | Decisions arise from a conjunction of factors, including perception, attention and learning processes, and individual characteristics. The strong link between visual stimuli and attentional mechanisms makes eye-tracking a powerful tool to provide a glimpse of what may occur at the brain level. Here, we aimed to explore the role of eye movements in value-based decision-making and to consider key substrates of the reinforcement learning theory. We analyzed eye data from two rhesus monkeys while performing a two-stage Markov decision task, known to elicit different learning strategies. By analyzing thousands of trials across dozens of sessions, we examined how gaze patterns influence behavior at a level of detail still not achievable in humans. Descriptive results of relevant ocular metrics, such as the number of saccades, meet the existing literature on binary choice, with mostly 1 or 2 saccades per decision stage. Results support a random first gaze, with no bias for the choice made or the choice of greatest value to the subject. On the other hand, the subject’s last look is a strong indicator of the chosen option. A Drift Diffusion Model approach established the baseline for gaze allocation and choice behavior association. Using Machine Learning, eye movement metrics alone showed considerable accuracy in predicting the upcoming choice. Adding the temporal factor via Recursive Neural Networks for forecasting proved to be beneficial. We conclude that visual perception and attention play a significant role in decisionmaking and are related to one’s learning processes. Our findings also highlight the benefits of gaze analysis for a thorough understanding of choice behavior.
|
|---|